Automatic Rule Creation
To access automatic rule creation, click Autodetect rules on the menu bar.
Important: Automatic rule creation is still under development. Review its limitations carefully before you rely on the generated rules.
Supported rules
Automatic rule creation currently supports only these rules: Paragraph, Heading, Block Paragraphs, Figure (raster and vectorial), and Table with its child rules Table Row, Table Data, Column Header, and Row Header. Other rules are not yet supported.
Rule organization
All generated rules for a page are placed in a Condition rule named Page followed by the page number, for example Page 3. Every page condition rule is placed inside a Region rule named Autodetect followed by a sequential count, for example Autodetect 1.
Reading order
Rules on a page are sorted top to bottom by Y position. Rules that fall on the same line, meaning they have close Y positions, are then sorted left to right by X position. This order might not match the intended reading order. Verify the reading order and reorder rules if needed.
Position and size values
Every generated rule has Export As set to Fixed for X, Y, width, and height. The root Region rule has these values set to Absent. As a result, if a page contains variable-length content, for example paragraphs or tables, the rules that follow that content might be positioned incorrectly on other documents. Adjust these positions manually.
Images
- Vectorial lines, based on maximum X/Y size, do not generate rules.
- Raster images with overlapping centers are merged into a single Figure rule.
- Intersecting raster images are merged into a single Figure rule when the overlapping area exceeds the configured X/Y tolerance.
Tables
Tables are generated first using the vectorial algorithm, based on line intersections, and then using the table-text detection algorithm. Depending on the drawing style, some tables might not be recognized correctly.
Table-text detection requires every cell in the table to contain text. Empty cells prevent recognition, and in some cases non-table text might be misidentified as a table.
A table is generated only when both the row count and column count are 2 or greater. The first row becomes Column Header cells. The remaining rows become Table Data cells.
Table-text detection identifies consecutive rows whose cell centers are aligned on the Y axis, within a configured tolerance.
Headings and alternative text
Only bold text chunks are detected as Heading rules. The feature does not set a heading level. Correct heading levels manually.
The feature does not generate alternative text for any rule. Make sure you add alternative text to figures, including logos or text-only images, since the feature cannot recognize text or meaning within an image.
Text detection
Text items on a page are merged into text chunks by proximity and font-size tolerance. These chunks are used for Paragraph, Heading, Table Data, and Block Paragraphs rules.
Processing scope and behavior
- Detection priority: tables using vectorial lines, tables using tabular text, Block Paragraphs, Paragraph and Heading, raster images, vectorial images.
- Pages with no text, image, or vectorial content do not get a generated Condition rule.
- Each page is processed independently. The feature does not reuse structures detected on other pages.
- The feature processes only the currently open PDF.
- Running automatic rule creation regenerates rules for the entire document each time. It does not account for existing rules or previously matched items. Delete rules from a previous run if you no longer need them.
- Rotated text, rotated content, or a rotated page is not supported.
Keyboard shortcut
The Autodetect rules option has no keyboard shortcut and cannot be configured in the application settings.